Nick Sharp

610 posts

Nick Sharp banner
Nick Sharp

Nick Sharp

@nmwsharp

3D geometry researcher: graphics, vision, 3D ML, etc | Senior Research Scientist @NVIDIA | running, hockey, baking, & cheesy sci fi | opinions my own | he/him

Seattle, WA Katılım Mayıs 2018
303 Takip Edilen6K Takipçiler
Sabitlenmiş Tweet
Nick Sharp
Nick Sharp@nmwsharp·
Giant new v1.2 release of Polyscope! Adds support for tet & hex meshes, transparency, ground shadows, slice planes, variable-size points, setting camera views, and more. Try Polyscope for easy 3D visualization in C++ & Python: polyscope.run Thread of new features! 1/n
English
3
49
331
0
Nick Sharp
Nick Sharp@nmwsharp·
@toshi2k2 @yang_yuezhi @CVPR Hi Prakhar, your paper looks very cool! To be clear our paper here is specifically focused on the geometric task of *convex decomposition* rather than semantic 3D segmentation; it's almost more of an acceleration structure. Looking forward to chatting more at CVPR!
English
0
0
0
74
Prakhar Kaushik
Prakhar Kaushik@toshi2k2·
@yang_yuezhi Great work. Would like to point out the first feedforward method for 3D part decomposition is ours. name-that-part.github.io. I shared this with @nmwsharp a month ago. Our method also works for non convex 3D parts. Will also be presenting this @CVPR , so can share ideas.
English
1
0
2
151
Yuezhi Yang
Yuezhi Yang@yang_yuezhi·
Excited to share our new work at CVPR 2026: Learning Convex Decomposition via Feature Fields. We introduce the first feedforward openworld model that generates high-quality convex decomposition for any 3D shapes in seconds, enabling faster simulation. 🔗research.nvidia.com/labs/sil/proje…
English
4
30
172
22.6K
Nick Sharp
Nick Sharp@nmwsharp·
Computing a convex decomposition of a shape is a classically-hard geometry problem, yet essential for fast physics simulators. Yuezhi found a way to accelerate it by training a large model!
Yuezhi Yang@yang_yuezhi

Excited to share our new work at CVPR 2026: Learning Convex Decomposition via Feature Fields. We introduce the first feedforward openworld model that generates high-quality convex decomposition for any 3D shapes in seconds, enabling faster simulation. 🔗research.nvidia.com/labs/sil/proje…

English
1
12
108
9.9K
Nick Sharp
Nick Sharp@nmwsharp·
@yongyuanxi Too long for a tweet :) All of the above pretty much! Personally I think most about tasks where we want to analyze or generate the real physical world around us in some way; fabrication is a great one. But this involves a stack of tools from basic data to high-level reasoning.
English
1
0
7
508
Towaki Takikawa / 瀧川永遠希
Towaki Takikawa / 瀧川永遠希@yongyuanxi·
@nmwsharp What do we mean exactly when we say “spatial learning”? (design for real world objects and systems? virtual spaces? navigating the world? or some abstract system like an LLM that is meant to be a primitive for compute?)
English
1
0
6
841
Nick Sharp
Nick Sharp@nmwsharp·
Totally agree. Spatial learning needs to be built with a base of tech that is scalable, efficient, precise, and controllable. 3D representations are exactly that. The best coding models are the ones that use tools the best. Care to guess what will distinguish the best 3D AI?
Ben Mildenhall@BenMildenhall

We don't expect LLMs to multiply numbers or sort lists directly within their output token stream. Instead, we ask them emit code and execute it in a separate runtime. Why predict the opposite outcome for simulating interactive worlds? worldlabs.ai/blog/3d-as-code

English
1
3
44
7.4K
Nick Sharp
Nick Sharp@nmwsharp·
@MenyJanos "IDEs" here is meant as more of an abbreviated blanket term for "code tooling", and I'd say that's gone into overdrive lately rather than disappearing. For me at least, LLM coding is filled with new interfaces, MCPs, review tools, etc, and my compiler is working harder than ever.
English
1
0
1
77
Janos
Janos@MenyJanos·
@nmwsharp It does seem like LLMs are killing IDEs though ..
English
1
0
0
60
Nick Sharp
Nick Sharp@nmwsharp·
Great read if you work in 3D. We're firmly in the middle of the "3D AI era"---why was this change so fast for text & pixels, but still so tricky for 3D? Representations are harder, and data doesn't exist. Yet in fields beyond vision, 3D is much more than means-to-an-end! (1/2)
Vincent Sitzmann@vincesitzmann

In my recent blog post, I argue that "vision" is only well-defined as part of perception-action loops, and that the conventional view of computer vision - mapping imagery to intermediate representations (3D, flow, segmentation...) is about to go away. vincentsitzmann.com/blog/bitter_le…

English
1
3
80
12.9K
Nick Sharp
Nick Sharp@nmwsharp·
@MattNiessner LLMs both help and hurt in this sense. In some ways they very much help us communicate clearly more efficiently. But also, they make it easy to disguise ill-posed ideas under a veneer of confident polish, which takes significant effort to detect.
English
0
0
1
336
Nick Sharp
Nick Sharp@nmwsharp·
@MattNiessner I disagree; the most important role of a paper is not any numerical score it achieves, but its ability to coherently communicate a new idea, so the reader can build something even better atop it. Clear presentation is hugely important. Papers are not high-score leaderboards!
English
2
0
6
372
Matthias Niessner
Matthias Niessner@MattNiessner·
Historically, academia used presentation quality as a proxy for scientific merit. Now that AI is eliminating polish overhead, everyone is confused, often stuck in debates whether we should allow LLMs. On the bright side, we are finally forced to evaluate the actual research content rather than extrapolating value from the text and visuals.
English
22
19
235
35.7K
Nick Sharp
Nick Sharp@nmwsharp·
@fchollet We considered this for discrete structured shape grammars, and identified key properties to make them act manifold-like under SGD. The "jump continuity" we discuss is this same cliff landscape---if you can make the cliffs smaller, descent works better! x.com/jackzzhang/sta…
Jack Zhang@jackzzhang

Can we apply gradient descent to discrete changes? In our new #SIGGRAPHAsia paper, we show that gradient descent can work on shape grammars, as in CAD and procedural modeling, but only if the grammars are designed correctly!

English
0
3
54
5.9K
François Chollet
François Chollet@fchollet·
Gradient descent is a powerful tool for optimization spaces that verify the manifold hypothesis, but the space of reasoning is discrete and combinatorial. GD fails in cliff-like landscapes where a single discrete change (a logical step) alters the entire outcome. Unless...???
English
102
58
963
118.1K
Nick Sharp
Nick Sharp@nmwsharp·
In engineering and art, geometry is often represented not as meshes or points, but as domain-specific structured *grammars*. In this work led by @milin_k_ and @jackzzhang, we investigated how to optimize these grammars ML-style with SGD. 4 simple rules make a huge difference!
Jack Zhang@jackzzhang

Can we apply gradient descent to discrete changes? In our new #SIGGRAPHAsia paper, we show that gradient descent can work on shape grammars, as in CAD and procedural modeling, but only if the grammars are designed correctly!

English
4
43
380
30.3K
Nick Sharp retweetledi
rishit dagli
rishit dagli@rishit_dagli·
📢want to produce realistic dynamic 3d worlds (with >100 splats) my new NVIDIA internship project, VoMP, is the first feed forward approach to convert input surface geometry to volumetric sim-ready assets by assigning real world physics materials 🌐Project: research.nvidia.com/labs/sil/proje… 📜Paper: arxiv.org/abs/2510.22975
English
2
18
95
9.7K
Nick Sharp
Nick Sharp@nmwsharp·
@swiftlysingh The majority of our interns are in PhD programs, but there can be opportunities for students at other stages as well so please apply!
English
0
0
1
457
Pushpinder Pal Singh
Pushpinder Pal Singh@swiftlysingh·
@nmwsharp This looks like a fantastic opportunity! Are you considering applications from Master's students?
English
1
0
0
579
Nick Sharp retweetledi
Masha Shugrina
Masha Shugrina@_shumash·
See your Gaussian Splats deform and collide under gravity! #NVIDIA Kaolin Library just released v0.18.0. github.com/NVIDIAGameWork… Join us at #SIGGRAPH tomorrow Sunday, Aug 10, Room 121-122 for a hands-on lab showcasing this and an intro to NVIDIA Warp, used under the hood.
English
0
3
23
2.2K
Nick Sharp retweetledi
Jake Rice
Jake Rice@TearsOfJake·
New at #SIGGRAPH2025: Can we make Perlin Noise stretch along some underlying vector field? Well it turns out it's possible with two simple additions to the original method! No need for advection or convolutions. Find the paper and implementations here: github.com/jakericedesign…
Jake Rice tweet media
English
5
33
198
16.8K
Nick Sharp
Nick Sharp@nmwsharp·
Also: this paper was recognized with a best paper award at SGP! Huge thanks to the organizers & congrats to the other awardees. I was super lucky to work with @yousufmsoliman on this one, he's truly the mastermind behind it all!
Nick Sharp@nmwsharp

Logarithmic maps are incredibly useful for algorithms on surfaces--they're local 2D coordinates centered at a given source. @yousufmsoliman and I found a better way to compute log maps w/ fast short-time heat flow in "The Affine Heat Method" presented @ SGP2025 today! 🧵

English
0
4
68
4.3K
Nick Sharp
Nick Sharp@nmwsharp·
We give two variants of the algorithm, and show use cases for many problems like averaging values on surfaces, decaling, and stroke-aligned parameterization. It even works on point clouds!
Nick Sharp tweet media
English
1
4
25
1.6K
Nick Sharp
Nick Sharp@nmwsharp·
Logarithmic maps are incredibly useful for algorithms on surfaces--they're local 2D coordinates centered at a given source. @yousufmsoliman and I found a better way to compute log maps w/ fast short-time heat flow in "The Affine Heat Method" presented @ SGP2025 today! 🧵
English
2
64
433
34.1K

Keşfet

@toshi2k2 @yang_yuezhi @CVPR @yongyuanxi @MenyJanos @MattNiessner @fchollet @milin_k_